Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
.. 2~'~~~~3
Blue Chip's Ref.: BM 1 WO ~lunchen, den
Yamaha's Ref.: H5123
Our Ref.: Pat 792136-95E
- 9, Jan. 1998
Translation of Patent Application PCT/EP94/03917
~ianai an~lvsis device having_at least one
strPtc-hed string_and one i~CkUt7
The invention relates to a signal analysis device having at least one
stretched
siring whose oscillatory length can be varied by resting it on at least one
fret, having
1 5 a pickup and having an evaluation device connected to the pickup.
A signal analysis device of this type can also be briefly designated as a
"guitar
synthesizer".
~ !n~ modern pop and rock music, there is an increasing trend no longer to use
~'- musical ~ instruments directly for the production of notes or sounds; but
just to
produce or analyze and convert electric signals, which are further processed
by
computers or other circuits. For this purpose, there are standardized
interfaces, of
which the MIDI interface has become relatively well known.
While a signal production or signal analysis of this type is accompanied by
relatively few difficulties in the case of keyboard musical instruments, since
in this
case each key has assigned to it precisely one pitch and the volume can, if
appropriate,
be determined via the speed of attack of the key, signal analysis in the case
of stringed
3 0 instruments, for example guitars, presents considerable difficulties. In
the case of
stringed instruments ~of this type, although each string has a fundamental
note
assigned to it, by pressing down the string on various pick-offs or frets, the
pitch of a
plucked, struck or otherwise excited string can be varied. In order to
determine the
CA 02174223 1999-11-22
2
correct pitch, it is necessary firstly to wait for the formation
of such a note and then to measure the frequency or duration of
at least one, but preferably more than one, period, in order to
be able to discover the pitch with the necessary reliability.
US Patent 4,823,667 therefore shows a signal analysis
device as an electronic musical instrument which is actuated in
the manner of a guitar, in which a frequency analyzer which
determines the frequency of the excited string is provided.
However, a procedure of this type leads to problems in relation
to time. In the case of a normal guitar, the lowest note has a
frequency of about 80 Hz (exactly: 82 Hz), so that one complete
oscillation needs a time of about 12.5 ms. Since, for reasons
of certainty, it is normally desired to measure two oscillations
in order to draw reliable conclusions, the time which is
necessary already adds up to 25 ms. In this case, it has not yet
been taken into account that the string still needs a certain
time after the excitation, for example by means of plucking or
striking, to pass into the steady state. For this purpose, as
a rule a non-negligible interval is likewise to be added, which
may well amount to twice a period length, so that the desired
pitch information is available only after 50 ms. A time delay
of 50 ms, however, is already distinctly noticeable to a
musician. It corresponds to the setting up of the loudspeaker
box at a distance of approximately 15 m.
As an alternative possible solution for this problem,
therefore, in US Patent 5,085,119 there are provided, on the neck
of the guitar switches which are activated when the corresponding
string is pressed down onto the desired fret. However, precisely
as in the case of a keyboard instrument, the pitch information
is no longer obtained by the string oscillation but by pressing
down a switch. This makes playing considerably more difficult.
The invention enables the obtaining of the pitch
information more rapidly in the case of a guitar synthesizer.
CA 02174223 1999-11-22
2a
More particularly, in the case of a signal analysis device
of the type mentioned at the outset in that the evaluation device
registers pulses or groups of pulses which run on the string past
the pickup after an excitation of the string and, on the basis
of the time sequence of the pulses or of the groups of pulses,
produces a signal which represents a pitch.
3'
One therefore no longer waits until an oscillation has been built up on the
string and is then measured, but rather evaluates so-called "plucking
transients",
that is to say pulses or pulse trains which result from the excitation of the
guitar
string. If a guitar string is plucked or struck, in the simplest case two
pulses or
traveling waves are produced, which move from the point of excitation in the
direction
of the clamping points of the string and, respectively, of the point where the
string is
pressed down onto the fret. There, they are reflected and run toward each
other once
more. After running to and fro several times, the known standing wave is then
formed,
which is normally responsible for the note production. However, it is also
possible to
1 0 measure the propagation time of these pulses on the string or to evaluate
it and, from
the propagation time or, respectively, the difference in propagation times
between
individual pulses, to obtain the necessary information about the string length
and
string tension and thus about the pitch. Of course, in actual fact individual
pulses do
not form, but rather groups of pulses. However, this does not change anything
1 5 concerning the principle on which the invention is based.
Preferably, the evaluation device also registers the polarity of the pulses or
groups of pulses and, from the time sequence of the pulses or groups of
pulses,
determines a signal which represents the position of excitation of the string.
The
2 0 position of excitation of the string, that is to say the position at which
the string is
plucked or struck or set into motion in another way, is one of the great
possible means
by which the player can use an individual style when playing the guitar.-
Since two
pulses or groups of pulses are available, which -move away from. the position
~-of
excitation in opposite directions on the string and are:reflected with_
corresponding -
2 5 time delays at the respective clamping positions- of the strings, on the
basis of the
different propagation times of the pulses it is also possible to obtain
information as to
where the position of excitation was located. This information is
obtainedvirtually as
rapidly as the information about the pitch, so that the. determination of the
position of.
excitation does not mean any further time delay.
The evaluation device preferably comprises a neural network which classifies
each sequence of pulses or groups of pulses into one of a multiplicity of
classes. The
sequences of pulses or groups of poises, which are to be assigned to a
specific pitch; in
each case have significant commonalities, which a neural network can discover -
3 5 relatively easily. It is possible here to be satisfied with similarities
between the.
individual pulse sequences or sequences of groups of pulses, without having to
evaluate
217223
4
each sequence of pulses exactly with respect tv time. The precise evaluation
with
respect to time can occasionally be accompanied by difficulties if the pulses
are not
present in the desired purity but are surrounded by disturbing noise. In this
case, it
can occasionally become difficult to define exact start and finish times for
the
allocation of the intervals of individual pulses or groups of pulses. On the
other hand,
a neural network can be programed in such a way that ii makes the decision as
to
which pitch is present and at which position the string has been excited
simply on the
basis of similarities. Here, a neural network has the advantage that it does
not
necessarily need explicitly specified rules in accordance with which it
assesses the
1 0 similarities. Rather, a neural network can be trained, that is to say, as
the result of
the presentation of a multiplicity of examples having the correct results, it
forms
algorithms or controlled responses for itself, which make it capable of
correctly
classifying subsequent examples. Furthermore, a neural network can to a
certain
extent make generalities, forming the rules for the generalities itself. The
neural
1 5 network is therefore in a position to detect sequences of pulses or
sequences of groups
of pulses relatively precisely even if the sequence of guises presented to it
does not
exactly coincide with an already trained sequence of pulses. Since neural
networks as a
rule are constructed with a multiplicity of processors operating in parallel,
they are
sufficiently fast to make the pitch signal available in the required short
time span.
It is also preferred that the evaluation device comprises a comparison device
which compares a pitch signal obtained from the siring in the steady state
with the
. signal obtained from the sequence of pulses and; in the event of a deviation
which. ,
exceeds a predetermined amount, triggers a-learning algorithm of the neural
network.
2 5 The evaluation device therefore does not restrict the pitch identification
to the
. _ evaluation of the "plucking transients". Rather, this evaluation is only
the beginning,
which nevertheless makes it possible- to make the pitch signal available
withinwthe
shortest time. The evaluation device also monitors whether the signal
detecied~ agrees
with the pitch which later builds up in- the oscillating string. If this is
so, the
3 0 "prediction" was correct and no further measures are. necessary. If the
"prediction" _
was wrong, however, there is a certain probability that the algorithm in
accordance
with which the neural network has assessed the similarity was erroneous: In
this
case, the result of the comparison can be used to make a further training
example
available to the neural network. On the basis of this training example, the
neural
3 5 network can learn anew and improve its identification algorithm.
5
Preferably, there is connected upstream of the neural network a selection
device which selects individual pulses from a group of pulses. This is
particularly of
advantage if the neural network provides only a restricted operating capacity.
In this
case, the quantity of information which the neural network must process can be
kept
smaller by means of a corresponding preselection.
A dedicated pickup is preferably provided for each string. This allows a
parallel sound signal production for each string to be realized, without
confusion of
the evaluation device being able to occur as a result of the plucking
transients, that is
1 0 to say the pulses running to and fro, which are different for all the
strings.
The invention is described in the following text with reference to a preferred
exemplary embodiment, in conjunction with the drawing, in which:
1 5 Fig. i shows a schematic representation of a signal analysis device,
Fig. 2 shows a schematic structure of a string and
Fig. 3 shows a schematic representation of a signal variation.
A signal analysis or production device 1 has six strings E1, H2; G3, D4, A5
and E6, v~rhich are strung in the manner of.a guitar. Provided. for each
string is a
pickup 2 which; for example, can be constructed. as an. electromagnetic. or
't ~? piezoelectric sound pickup. The pickups ~2 are connected to an
analogldigital converter
2 5 3 which, in the exemplary embodiment shown; has one channel for each
pickup 2, that
is to say is designed with six channels.
The analog/digital converter 3 is connected to a microprocessor 4 which
provides the input and output management for a neural network 5. A selection
device 6
3 0 can also be provided between the microprocessor 4 and the- neural network
5, the
function of which selection device will be described later.
Moreover the analogldigital converter 3 is connected to a frequency meter 7:
The frequency meter 7 and the microprocessor 4 are connected to a comparison
device
3 5 8. The comparison device 8 is connected to a MIDI interface 9. The
comparison device
8 is likewise connected to the neural network 5, to be specific to a learning
input 10.
-o
6
Under the management of the microprocessor 4 and, if appropriate,
conditioned by the selection device 6, the neural network 5 receives a
sequence of
pulses or groups of pulses and classifies these sequences in each case into
one of a
multiplicity of specific classes. Here, each class allows a conclusion as to
the pitch
and, if appropriate, also as to the position of excitation of the string, as
will be .
explained in the following text.
Fig. 2 shows schematically a string 11 which is strung between a fixed
1 0 clamping point 12 and a clamping point 13 at which the tension can be set.
The siring
11 stretches over a guitar neck 14 on which there are arranged various frets
15.
Shown by an arrow 16 is one fret, on which the siring 11 is pressed down. This
fret
16, togeiher~with the clamping' point 12, determines the effective length of
the string
11. The pertinent pickup 2 is arranged under the string.
By means of a triangle 17, which is intended to symbolize a plectrum or a
similar plucking implement, a position of excitation for the string 11 is
shown. If the
siring 11 is now plucked or struck at this position of excitation, a standing
wave of
the frequency which is characteristic of the pitch is not established
directly. Rather, a
2 0 transient process begins, which can be described in a simplified way by
saying that
two pulses 18, 19 run to the left and to the right from the position of
excitation. These
pulses or traveling waves are differentiated. from each other .by a drawn-in 1
and a
drawn-in 2. The pulse 78 now. runs to- the. left as far. as the fret.16, .on
which the
y .y string is held down. There it is reflected, with phase reversal, and runs
back once .
more. In the same way the pulse 19 runs to the right as far.as the clamping
port 12,
where it is reflected, with phase reversal, and runs back once more. The
pulses or
waves, running to and fro; overlay one another- and after a short time form
the known
standing wave with which the string 11 oscillates.
3 0 However, the pulses 18, 19 run past the pickup 2. A corresponding time
diagram is shown in Fig. 3. It can be seen here that the first pulse, which
is. intended
to have a positive amplitude, crosses the pickup at a time t1, while its
reflection, now
having a negative amplitude, crosses the pickup at a time t2. At a time t3,
the second
pulse, reflected at the clamping point 12, reaches the pickup, while it runs
over the
3 5 pickup 2 once more at a time t4. This is then the second pulse reflected
for the second
time, specifically at the fret 16. At the times t5 and t6, the first pulse,
which has
2174223
7
then been reflected at the clamping point 12 and at the fret 16, runs once
more over
the pickup 2 and, at the times t7 and t8, the second pulse, which has then
been
reflected once more at the clamping point 12 and at the fret 16, runs over the
pickup
2.
The velocity of motion or traveling velocity of the pulses 18 or 19 on the
siring 11 is known. The active length of the string 11 can now be determined
from the
time difference T1, which is the difference between the times t5 and t1, with
the aid
of this traveling velocity. However, this is also the length which is
responsible for the
1 0 pitch of the string 11. Providing that the distance of the pickup 2. from
the fret 16 and
from the frets 15 is known, the distance T2, that is to say the interval
between the
times t2 and t1, would in principle also be sufficient. However, this
relinquishes the
possibility of fine tuning, since the guitarist has the possibility of varying
the pitch
by means of slight displacements of his finger on the frets 15,16. Moreover,
the
1 5 pulses cannot be distinguished so clearly in many cases, as is shown in
Fig. 3 for
reasons of simplicity. Rather, blurring of the individual pulses can occur, in
particular if, when the string 11 is plucked or struck, individual pulses, as
shown,
are not produced, but rather whole groups of pulses.
2 0 However, in almost alf cases, conclusions can be drawn as to the position
of the
excitation from the time difference T3, that is to say from the difference
between the -.
times t3 and t1. lf, from the difference T1, the string length is knowri, it
is- possible
to calculatev back from the difference T3 to find at- which fraction of the
siring the
excitation has taken place.
Nevertheless; the measurement of time for determining the interval between
the pulses shown is occasionally subject to uncertainties. For this reason,
using the
selection device 6, individual pulses are selected from the sequence of groups
of pulses
which are registered by the pickups 2 and. said individual pulses are fed to.
the neural
3 0 network 5. The neural network can identify similarities between individual
sequences
of groups of pulses and classify the "plucking transients", which are
represented by
these sequences of pulses, in such a way that their assignment to individual
classes,
which in each case reproduce a pitch and a position of excitation, is possible
with
great certainty. The identification sequence is triggered here by the
occurring pulses.
3 5 The successive positive and negative pulses or groups of pulses are
forwarded to the
neural network, which tries on each occasion to assign the pattern picked up
or the
.. X174223
8
sequence picked up to a previously learned sequence. This detection sequence
is
repeated until either the neural network has produced a positive result or the
frequency meter 7 has provided the corresponding information. If the neural
network
is still in the learning or training phase, in many cases the frequency meter
will be
quicker. However, after a certain training phase, the neural network 5, which
can
itself form the rules for the identification if it is programed accordingly,
has stored
sufficient information to be able to undertake the classification itself in an
extraordinarily effective manner. The neural network 5 also forms specific
rules for
generalities, so that even patterns which have not been learned specifically
can be
1 0 identified, providing these have specific similarities to the examples
already learned.
Since the frequency meter undertakes a pitch identification in parallel,
further learning is also possible during the operation of the signal analysis
device 1.
The comparison device 8 compares the pitch determined by the neural network 5
with
1 5 one determined later by the frequency meter 7. Here, it is possible on the
one hand to
follow the fine pitch changes, which are a means of expression of the player,
on the
other hand, using this procedure, errors or inaccuracies in the algorithm
which is
applied by the neural network 5 can be discovered and eliminated. The
comparison
device 8 specifically couples the determined error back into the neural
network 5 and
2 0 triggers a new learning algorithm, so that the same error cannot occur
again, as a
result of the improved identification capability. !n the event that no
difference occurs,
the comparison device 8 forwards the signal or signals unchanged to the MIDI
interface 9.
2 5 The output results of the neural network are processed further in such a
way
that the MIDI interface 9 can make MIDI signals available, which can drive a
MIDI
synthesizer or an expander module. The pitch encoded in the MIDI signal
corresponds
in this case to the pitch of the guitar string. Moreover, the plucking
position can also
be captained in the MIDI signal as monitoring information, as~ an encoded
sound quality
3 0 character.
